CNN Based Deep Learning for Vehicle Re-identification

Vehicle re-identification is a finding of wanted vehicle from a large amount of vehicles’ image. This one of basic problem in computer vision, that a re-identification system uses appearance, illumination and viewpoint (i.e. pose and orientation) of vehicle for recognition purpose. This paper presents the design of a deep learning structure, based on convolutional neural networks (CNN) for vehicle detection and identification of vehicle. The single-shot multibox detector (SSD) algorithm that is based on CNN is applied for detection of vehicles, and then the CNN structure is applied for identification of the detected vehicles. The combination of SSD and CNN structures are presented for accurate detection and identification of vehicles The designed system uses global and local deep features and identify the vehicles in a short time with high accuracy. The proposed framework performs the state-of-art performance in a realistic publicly available dataset, namely VeRi-776 dataset. The accuracy of identification of proposed system is achieved as 99.3%. The obtained results prove the effectiveness of the proposed algorithm in vehicle re-identification.